Anton likes the suggestion of adding some monitoring stuff. Ideally a job that every X (e.g. 20) seconds checks size of the queue and logs some warnings if the queue is bigger than Y (e.g. 100). If the queue is bigger than Y for Z (e.g. 10) minutes then send a thread dump to the logs as well.

It would be good if X, Y, Z are system properties (maybe defined the Spring way).

That way if customers do start having slowness problems we should be able to get some decent information as to whats causing it and the best possible ways to fix it.